Quote:
Originally Posted by
malmentalÂ

funny, I was coming back to correct myself..
http://www.asus.com/Motherboards/Intel_Socket_1155/P8P67_WS_Revolution/#specifications
Multi-GPU Support Supports NVIDIA® 3-Way SLI™ Technology
Supports NVIDIA® SLI™ Technology
Supports AMD Quad-GPU CrossFireXâ„¢ Technology
Supports AMD CrossFireXâ„¢ Technology
CUDA Core CUDA support:
Up to 4 NVIDIA Tesla GPUs
Expansion Slots 2 x PCIe 2.0 x16 (x16, x8)
2 x PCIe 2.0 x16 (x8 mode)
3 x PCIe 2.0 x1
compromise and say it's x16/x8 for the SLi (dual)
Not sure what's going on here but two slot config runs x16/x16 pcie 2.0. Four slots filled is x8/x8/x8/x8. 3 slots filled is x16/x/8/x8. Nvidia drivers are limited to trisli, it's a software thing and not hardware. Anyways, any of the nf200 z68/p67 boards can do x16/x16. However, running only two cards, you'd be hard pressed to saturate the lanes.
OP there was never a quadfire board made on z68. Latencies are overblown on nf200 and
Quote:
the NF200 adds extra latency and the performance of 2x16 is worse than 2x8
is just all wrong.
If the OP is looking to go quad, skip z77 (G1 Sniper 3 or p8z77 ws) and get a x79 ud3/x79 formula/x79 RIVE with a 3820. That way you get 40 lanes off the cpu without having to resort to using a multiplexer like the PEX8727, never mind that it's an awesome bridge chip in itself. Regardless the PEX8747 in the z77 quad boards are a bridge and not off the cpu like the x79 boards. +1 for the x79 boards obviously.
Edited by tsm106 - 5/25/12 at 11:38pm